JP2985976B2 - Syllable recognition device with tongue movement detection - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a tongue movement detection type syllable recognition device which detects a utterance of a user (speaker) or a tongue movement accompanying a utterance movement (mouth operation of utterance) to recognize a syllable. It is.

[0002]

2. Description of the Related Art A syllable recognizing apparatus for recognizing a syllable uttered by a user for the purpose of inputting information by voice to a computer or the like has been developed.

As a syllable recognition device of this type, for example, the following device has been proposed.

(1) A syllable recognition system which analyzes a speech waveform detected by a microphone by DP matching (Sakoe
Et al., Journal of the Acoustical Society of Japan, 2, 9, p. 43-49, 197
8).

(2) A syllable recognition system which analyzes a speech waveform detected by a microphone using a hidden Markov model (R)
abiner. L. R. et al, Bell Syst.
Tech. J. , 62, 4, p. 1075-1105,
1983).

(3) A syllable recognition method in which a speech waveform detected by a microphone is analyzed by a neural network (Kawahara et al.,
IEICE, Speech Technical Report SP88-31, 198
8). However, these syllable recognition devices require a quiet environment because the recognition rate of syllables decreases when noise is mixed in the voice waveform detected by the microphone. There is a drawback that recognition cannot be performed unless voice is emitted.

In order to solve this drawback, a syllable recognizing device which uses information relating to utterance other than speech in a complementary manner has been considered, and as one of them, a syllable recognizing device which performs lip reading by a computer has been proposed ( Peta
jan E. , IEEE CVPR'85, p. 40-47, 1985).

[0008]

However, the above-described syllable recognition apparatus for performing lip reading by a computer uses the camera to detect the movement of the user's lips to perform the lip reading. There is a drawback that the usable place is restricted because it needs to be installed in front of the camera.

The object of the present invention is not limited to the place of use,
Another object of the present invention is to provide a syllable recognition device for detecting tongue movement that can be used even in a noisy environment.

[0010]

The tongue movement detection type syllable recognition apparatus of the present invention comprises a tongue movement detection transmission unit installed in the oral cavity and a syllable recognition unit installed outside the oral cavity. A light-emitting element that irradiates near-infrared light, a light-receiving element that receives reflected light of the near-infrared light and converts the light into a light-receiving signal, and transmits the light-receiving signal converted by the light-receiving element to a transmission signal. And a battery that supplies power to the transmitter, the light-emitting element, and the light-receiving element, and the syllable recognition section receives the transmission signal. A receiver for demodulating the received light signal from the transmission signal received by the antenna, a neural network for outputting an output value of each output unit according to the received light signal demodulated by the receiver, Of each of the output units output from the neural network Performing a magnitude comparison of force values, the output value and a comparator for outputting a syllable corresponding to said output unit is maximum.

[0011]

In the tongue movement detection type syllable recognition device of the present invention, the utterance or utterance movement of the user obtained by detecting the reflected light of the near-infrared light emitted from the light emitting element toward the tongue surface by the light receiving element. Using the received light signal according to the accompanying tongue movement, the syllable recognition is performed by the neural network and the comparator, so that the voice recognition can be performed without detecting the voice emitted by the user.
Syllables can be accurately recognized even in a noisy environment or an environment where absolute silence is required. Also,
Since the received light signal is converted to a transmission signal and transmitted from the tongue movement detection transmission unit to the syllable recognition unit, the syllable recognition unit can be installed at any location as long as the transmission signal can be received,
The place of use is not restricted.

A syllable recognizing unit for detecting a voice emitted during the learning operation and converting the voice into a detected voice signal; an amplifier for amplifying the detected voice signal converted by the microphone; A speech recognition circuit for recognizing a syllable from the speech signal and outputting a teacher signal pattern indicating the syllable, and a neural network, wherein the teacher signal pattern input from the speech recognition circuit and the receiver at the time of learning operation. By further providing a learning function for learning the relationship between the received light signal and the syllable from the demodulated received light signal, the learning operation can be performed for each user, and then the recognition operation can be performed.

[0013]

Next, embodiments of the present invention will be described with reference to the drawings.

FIGS. 1A and 1B are block diagrams showing a first embodiment of a syllable recognition device of the tongue movement detecting type according to the present invention, and FIGS.
FIG. 1A is a diagram showing the appearance of the storage container 15 of the tongue movement detection / transmission unit 10 of FIG. 1A, and FIG.
FIG. 4 is a diagram illustrating a method of setting the neural network in the oral cavity, and FIG. 4 is a diagram illustrating a configuration of the neural network 26 in FIG.

The syllable recognition device of the tongue movement detecting type of this embodiment
It comprises a tongue movement detection transmission unit 10 installed in the oral cavity and a syllable recognition unit 20 installed outside the oral cavity.

As shown in FIG. 1A, the tongue movement detecting and transmitting section 10 includes a plurality of light emitting elements 11 for emitting near-infrared light L.
And the received light signals α, β, γ
A plurality of light-receiving elements 12 for converting the light-receiving signals α, β, and γ converted by the light-receiving elements 12 into transmission signals W and transmitting the transmission signals W to the syllable recognition unit 20; It comprises a light emitting element 11 and a battery 14 for supplying power to each light receiving element 12.

As shown in FIG. 1B, the syllable recognition unit 20 includes an antenna 21 for receiving the transmission signal W, and an antenna 2
The receiver 22 demodulates the received light signals α, β, and γ from the transmission signal W received in step 1, the microphone 23 that detects the sound emitted during the learning operation and converts the sound into the detected sound signal S, and the microphone 23. Amplifier 2 for amplifying converted detection audio signal S
4, a speech recognition circuit 25 for recognizing a syllable from the detected speech signal S amplified by the amplifier 24 and outputting a teacher signal pattern indicating the syllable, a teacher signal pattern input from the speech recognition circuit 25, and a learning operation. The received light signals α, β, γ are demodulated by the receiver 22 at
In addition to having a learning function of learning the relationship between γ and syllables, the output values y _{1 to} y _N of the output units 53 _{1 to} 53 _N (depending on the received light signals α, β, γ demodulated by the receiver 22) 4) and a neural network 26 for outputting
Output value y of each output unit 53 _{1 to} 53 _N output from
And a comparator 27 for comparing the magnitudes of _{1 to} y _N and outputting a syllable corresponding to the output unit having the maximum output value.

The tongue operation transmitter 10 is stored in a storage container 15 shown in FIG. 2, and the tongue operation transmitter 10 is installed in the oral cavity as shown in FIG. Two Y-shaped bridges 16 provided near both ends are
This is done by being inserted between each tooth.
At this time, the light receiving element 11 irradiates the near infrared light L toward the tongue surface 19, and the light receiving element 12 can receive the reflected light of the near infrared light L from the tongue surface 19. Fifteen
Is installed on the upper jaw such that the light emitting element 11 and the light receiving element 12 face the tongue surface 19.

Further, as shown in FIG. 4, the neural network 26 samples the light receiving signals α, β, γ with sample values d _{1 to} d ₅ , e _1.
~e _5, f ₁ ~f ₅ (see FIG. 7) is input as numerical pattern, an input layer 31 comprising a plurality of input units 51 ₁ to 51 _15, via a first link 36 having a weight 39 Layer 32 composed of a plurality of intermediate units 52 _{1 to} 52 _M connected to the input layer 31 and a plurality of outputs connected to the intermediate layer 32 via a second link 37 having a weight 39. Output layer 33 composed of units 53 _{1 to} 53 _N
And sample values d _{1 to} d ₅ , e _{1 of} the respective light receiving signals α, β, γ
To e _5, f ₁ ~f numerical pattern of _five, the teacher signal pattern input from the output unit 53 ₁ to 53 _N of the output value y ₁ consisting ~y _N output pattern and a voice recognition circuit 25 of the output layer 33 And the first and second links 3
And a weight control unit 34 for updating each of the weights 39 of 37. Each output module 53 ₁ to 53 _N is arranged to correspond to each monosyllabic (e.g., the output unit 53 ₁ is a single syllable "Ah", the output unit 53 ₂ is the monosyllable "have", output unit 53 ₃ corresponds to a single syllable "cormorant").

Next, the operation of the syllable recognition apparatus of the tongue movement detection type according to the present embodiment will be described separately for a learning operation and a recognition operation.

First, the learning operation will be described with reference to a flowchart shown in FIG. 5, a diagram showing the flow of each signal shown in FIG. 6, a waveform diagram shown in FIG. 7, and a waveform diagram shown in FIG.
Note that, for simplicity, the number of light emitting elements 11 and light receiving elements 12 will be described as three.

The learning operation is performed in a quiet environment in which the voice recognition circuit 25 can correctly recognize syllables.
In this environment, a voice composed of a single syllable is emitted by the user 40 (step 110). The emitted sound is detected by the microphone 23 and converted into a detected sound signal S. After the detected voice signal S is amplified by the amplifier 24, the voice recognition circuit 2
5 is input. The speech recognition circuit 25 performs speech recognition based on the detected speech signal S, thereby recognizing what the uttered monosyllable was (step 111).
). On the other hand, the operation of the tongue when a sound consisting of a single syllable is emitted is such that the near-infrared light L is emitted from the light emitting element 11 toward the tongue surface 19 and the reflected light of the near-infrared light L from the tongue surface 19 Are detected by being received by the light receiving element 12, and are converted into light receiving signals α, β, and γ as shown in FIGS. 7A to 7C, for example. Each of the received light signals α, β, and γ is converted into a transmission signal W by the transmitter 12 and then transmitted to the syllable recognition unit 2.
0 (step 112). The light emitting element 1
1, power required to operate the light receiving element 12 and the transmitter 12 is supplied from the battery 13 to each. The transmission signal W is received by the antenna 21 of the syllable recognizing unit 20 and demodulated by the receiver 22, so that each of the light receiving signals α,
After being returned to β and γ, they are input to the neural network 26.

Each received light signal α,
Extraction of β and γ is performed as follows (step 113).

By performing full-wave rectification on the detected audio signal S input from the amplifier 24 shown in FIG. 8A, an envelope waveform E shown in FIG. 8B is obtained. Thereafter, times t _L1 and t _{L2 at} which the amplitude of the envelope waveform E reaches the predetermined threshold value θ ₁ are obtained, and a gate signal whose amplitude is “1” from time t _L1 to time t _L2 shown in FIG. Create G. With the gate signal G, the received light signal α input from the receiver 22 is applied to an observation window as shown in FIG. Extraction of the other two light receiving signals β and γ is performed in the same manner.

Each of the light-receiving signals α,
β and γ are sampled at predetermined time intervals as shown in FIGS. 7A to 7C, and each sample value d ₁ to
d ₅ , e _{1 to} e ₅ , f _{1 to} f ₅ are input patterns 51 _{1 to} 51 of the input layer 31 of the neural network 26 as numerical patterns.
₁₅ are input to each (step 114). on the other hand,
The monosyllable recognized by the speech recognition circuit 25 is input to the weight control unit 34 of the neural network 26 as a teacher signal pattern for learning. For example, when the single syllable is “U”, it is input to the weight control unit 34 of the neural network 26 as the learning teacher signal pattern “00100... 0”.

The neural network 26 calculates each sample value d _{1-
d 5, e 1 ~e 5,} f 1 ~f of _five numeric pattern and a known error backpropagation the relationship between the teacher signal pattern for learning of the recognized syllable in the speech recognition circuit 25 (D.E Rum
elhart et al., Parallel Distr
ibued Processing, MIT Pre
ss. , 1986) (step 11).
Five ). This learning is performed, for example, with the teacher signal pattern “00”.
100 "for a single syllable of the output values y ₁ ~y _N of the output units 53 ₁ to 53 _N of the neural network 26" ... 0 output value y of the output unit 53 ₃ corresponding to the Hare " ₃ until the monosyllable “U” is output from the comparator 27, the first and second links 36, 3
This is performed by repeating the updating of the weight 39 of step 7 (step 116).

When the learning of one single syllable is completed in this way, the learning of the other single syllable is performed in the same manner. When the learning of all the single syllables is completed, the learning operation is completed (step 117). .

Next, the recognition operation of the tongue movement detection type syllable recognition device of the present embodiment will be described with reference to the flowchart shown in FIG. 9, the diagram showing the flow of each signal shown in FIG. 10, and the waveform diagram shown in FIG. explain.

In a noisy environment, the user 40 does not necessarily need to emit a voice consisting of a plurality of monosyllables, but only needs to perform a utterance operation (mouth operation of utterance) (step 210).
). The near-infrared light L is emitted from the light-emitting element 11 toward the tongue surface 19, and the reflected light of the near-infrared light L from the tongue surface 19 is received by the light-receiving element 12, so that the user 40 utters or speaks. Of the tongue associated with the light is detected and converted into light receiving signals α, β, and γ. Each of the received light signals α, β, and γ is converted into a transmission signal W by the transmitter 12 and then transmitted to the syllable recognition unit 2.
0 (step 211). The transmission signal W is received by the antenna 21 of the syllable recognition unit 20 and demodulated by the receiver 22 to be returned to the respective light reception signals α, β, and γ, and then input to the neural network 26. During the recognition operation,
Since the user 40 does not always speak, the neural network 2
6, the light receiving signals α, β, and γ cannot be sampled using the detected voice signal S as in the learning operation.
Therefore, paying attention to the fact that the noise superimposed on each of the light receiving signals α, β, and γ does not greatly differ from that in a quiet environment even under a noise environment, as shown in FIG. From the time t _{K1 at} which the amplitude of the light receiving signals α, β, γ first crosses the predetermined threshold θ ₂ , an observation window of a predetermined time width T _W is determined,
By applying this observation window to each of the received light signals α, β, and γ, the received light signals α, β, and γ are extracted (step 212). Each of the received light signals α, β, and γ thus applied to the observation window is sampled at predetermined time intervals,
Each sample value _{d 1 ~d 5, e 1 ~e} 5, f 1 ~f 5 is input as a numerical pattern to the input layer 31 of the neural network 26 (step 213). Output values of the output units 53 ₁ to 53 _N of the output layer 33 of the neural network 26 y ₁ ~y _N is the comparator 2
7, each of the output values y _{1 to} y _{N is} output from the comparator 27.
Are compared, and single syllables corresponding to the output unit having the maximum output value are sequentially output from the comparator 27 (step 214). The above operation is repeated until all the recognition operations are completed (step 215).

Therefore, in the tongue movement detection type syllable recognition apparatus of the present embodiment, the recognition operation can be performed without using the voice uttered by the user 40, so that the syllable recognition can be performed accurately even in a noise environment. In addition, since the tongue movement detection and transmission unit 10 installed in the mouth of the user 40 converts each of the received light signals α, β, and γ into a transmission signal W and transmits the transmission signal W to the syllable recognition unit 20, the installation location of the syllable recognition unit 20 Is not limited.

The extraction of the light receiving signals α, β, and γ in the neural network 26 is not limited to that shown in FIG.
For example, it may be performed as shown in FIG. That is, after sampling each of the light receiving signals α, β, γ at predetermined time intervals and storing them in the memory, the respective sample values of the acceleration signals α, β, γ having a predetermined time width from an arbitrary time t ₀ are obtained. By reading from the memory, each light receiving signal α, β,
γ is multiplied by a first observation window, and then, from time t ₀ + Δt, each sample value of each of the light receiving signals α, β, and γ having the predetermined time width is read from the memory, so that each of the light receiving signals α,
A second observation window is applied to β and γ, and then the time t ₀ + 2 · Δ
A third observation window is applied to each light receiving signal α, β, γ by reading each sample value of each light receiving signal α, β, γ of the predetermined time width from t from t. The light receiving signals α, β, and γ may be extracted by repeating the above operation a predetermined number of times. However, in this case, for example, each of the light receiving signals α,
Since there are no single syllables corresponding to β and γ, the output layer 33 of the neural network 26 is
It is better to add an output unit that indicates "no syllable" to.

Further, in the syllable recognition apparatus of the tongue movement detection type shown in FIG. 1, the recognition operation is performed using only the respective light receiving signals α, β, and γ. If the detection can be performed to some extent by the microphone 23, the recognition operation may be performed using the detected audio signal S complementarily as shown in FIG. That is, in this case, the detected voice signal S is input to the input layer 31 of the neural network 26 shown in FIG.
, A plurality of input units to which the respective sample values are respectively input are added, and each of the received light signals α, β, γ and the detected audio signal S
After performing a learning operation similar to the above-described learning operation using the learning signal and the teacher signal pattern, a recognition operation similar to the above-described recognition operation is performed using each of the light receiving signals α, β, and γ and the detected voice signal S. Is also good.

In the learning operation, learning was performed for each single syllable. However, the number of input units of the input layer 31 of the neural network 26, the number of intermediate units of the intermediate layer 32, and the number of output units of the output layer 33 are different. And the learning may be performed for each of a plurality of single syllables.

As a method of learning a single syllable, a method of learning by a back propagation method using a neural network 26 composed of a multilayer neural network for recognizing a spatiotemporal pattern as shown in FIG. But, for example,
IEICE, Speech Technical Report, SP87-100, 19
A time-delay neural network (TDNN: Phoneme recognize, described in November 1987)
ion using time-delay ne
Other neural networks that process spatiotemporal patterns, such as ural networks, may be used.

FIG. 14 is a block diagram of a syllable recognizing unit 70 showing a second embodiment of the syllable recognition type syllable recognition apparatus of the present invention.

The tongue movement detection type syllable recognition device of this embodiment
The difference from the tongue movement detection type syllable recognition device shown in FIG. 1 is that a neural network 76 that has been subjected to a learning operation in advance so as to be able to handle an unspecified number of users is used. Therefore, in the tongue movement detection type syllable recognition device of the present embodiment, since the above-described learning operation is unnecessary, the syllable recognition unit 70 includes:
A microphone for detecting a voice emitted by the user and converting it to a detected voice signal S, an amplifier for amplifying the detected voice signal S, and a voice recognition circuit for outputting a teacher signal pattern during a learning operation are not required. Note that the recognition operation in the tongue movement detection type syllable recognition device of this embodiment is performed in the same manner according to the flowchart shown in FIG.

In the above description, each light receiving signal α,
The sampling of β and γ was performed by the neural network 26, but the receiver 22 may have this function, or a sampling device may be separately provided. Further, as shown in FIG. 3, the tongue movement detection transmitting unit 10 is installed with the two bridges 16 inserted between the teeth, respectively. However, for example, it is also possible to connect or integrate with the dentures It is.

The application fields of the tongue movement detecting type syllable recognition device of the present invention include, in addition to a voice input device to a computer,
It can be applied to a word processor, a machine translator, a telephone number of a car phone, a voice input device to a computer loaded in a vehicle and a computer loaded in an aircraft cockpit, and an application as a secret talk tool by networking a syllable recognition unit.

[0039]

As described above, the present invention has the following effects.

(1) A light emitting element and a light receiving element provided in the tongue movement detection transmitting unit detect a user's utterance or a tongue movement associated with the utterance movement and convert it into a light reception signal.
By transmitting the received light signal to a syllable recognizing unit provided with a neural network and a comparator, the syllable is recognized using the received light signal. Therefore, in a noise environment or an environment where absolute silence is required. Syllables can be recognized accurately even below. Further, the syllable recognition unit may be installed in any location as long as the transmission signal can be received, so that the use location is not limited.

(2) By making the neural network have a learning function, a learning operation can be performed for each user, and then a recognition operation can be performed. Therefore, a recognition operation can be performed accurately for any user. it can.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a first embodiment of a tongue movement detection type syllable recognition device according to the present invention, in which (A) is a block diagram of a tongue movement detection and transmission unit, and (B) is its syllable recognition. It is a block diagram of a part.

FIG. 2 is a diagram illustrating an appearance of a storage container 15 of the tongue movement detection transmission unit 10 illustrated in FIG.

FIG. 3 is a diagram showing a method of installing the tongue movement detection transmitting unit 10 shown in FIG.

FIG. 4 is a diagram showing a configuration of a neural network 26 shown in FIG. 1 (B).

FIG. 5 is a flowchart illustrating a learning operation of the tongue motion detection type syllable recognition device shown in FIG. 1;

6 is a diagram showing a flow of each signal for explaining a learning operation of the tongue movement detection type syllable recognition device shown in FIG. 1;

FIGS. 7A and 7B are waveform diagrams showing each light receiving signal, FIG. 7A is a waveform diagram of a light receiving signal α, FIG. 7B is a waveform diagram of a light receiving signal β, and FIG.

8A and 8B are waveform diagrams for explaining a method of extracting each light receiving signal at the time of a learning operation in the neural network 26 shown in FIG. 1B, wherein FIG. 8A is a waveform diagram of a detected voice signal S, and FIG. FIG. 7C is a waveform diagram of the envelope waveform E, FIG.
(D) is a waveform diagram of the light reception signal α.

FIG. 9 is a flowchart illustrating a recognition operation of the tongue movement detection type syllable recognition device illustrated in FIG. 1;

10 is a diagram showing a flow of each signal for explaining a recognition operation of the tongue movement detection type syllable recognition device shown in FIG. 1;

11 is a waveform diagram illustrating a method of extracting each light receiving signal during a recognition operation in the neural network 26 illustrated in FIG. 1 (B).

12 is a waveform diagram illustrating another method of extracting each light receiving signal during a recognition operation in the neural network 26 illustrated in FIG. 1 (B).

13 is a diagram showing a flow of each signal for explaining another recognition operation of the tongue movement detection type syllable recognition device shown in FIG.

FIG. 14 is a block diagram of a syllable recognition unit according to a second embodiment of the tongue movement detection type syllable recognition device of the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 10 Tongue motion detection transmission part 11 Light emitting element 12 Light receiving element 13 Transmitter 14 Battery 15 Storage container 16 Bridge 19 Tongue surface 20, 70 Syllable recognition part 21, 71 Antenna 22, 72 Receiver 23 Microphone 24 Amplifier 25 Voice recognition circuit 26, 76 neural network 27, 77 comparator 31 input layer 32 intermediate layer 33 output layer 34 weight control unit 36, 37 link 39 weight 51 _{1 to} 51 ₁₅ input unit 52 _{1 to} 52 _M intermediate unit 53 _{1 to} 53 _N output unit α , Β, γ Light reception signal S Detection voice signal W Transmission signal d _{1 to} d ₅ , e _{1 to} e ₅ , f _{1 to} f ₅ Sample value y _{1 to} y _N output value

────────────────────────────────────────────────── ─── Continued on the front page (51) Int.Cl. ⁶ Identification code FI G10L 3/00 571 H04B 7/00 G01S 17/88 Z H04B 7/00 A61B 5/10 310L (72) Inventor Kazuhiko Shinozawa Tokyo Nippon Telegraph and Telephone Corporation, 1-6-1, Uchisaiwai-cho, Chiyoda-ku (56) References JP-A-2-297599 (JP, A) JP-A-2-144036 (JP, A) JP-A-2-196537 (JP) JP-A-57-209036 (JP, A) JP-A-62-174787 (JP, A) JP-A-2-144035 (JP, A) JP-A-2-140146 (JP, A) 58-79297 (JP, A) JP-B 3-6519 (JP, B2) JP-B 61-61,760 (JP, B2) JP-B 3-502770 (JP, A) 1989 IEICE Autumn National Convention Proceedings, 6th volume, “Syllable utterances Study of recognition by the neural network of the previous preparation potential space-time pattern ", p. 6-8, (1989) (58) Field surveyed (Int. Cl. ⁶ , DB name) G10L 3/00-9/20 JICST file (JOIS)

Claims (2)

(57) [Claims] 1. A light emitting element comprising a tongue movement detection transmission unit installed in the oral cavity and a syllable recognition unit installed outside the oral cavity, wherein the tongue movement detection transmission unit irradiates near-infrared light; A light-receiving element that receives the reflected light of the near-infrared light and converts it into a light-receiving signal, and a transmitter that converts the light-receiving signal converted by the light-receiving element into a transmission signal and transmits the signal to the syllable recognition unit, A battery that supplies power to the transmitter, the light emitting element, and the light receiving element, wherein the syllable recognition unit receives the transmission signal, and receives the light reception signal from the transmission signal received by the antenna. A receiver for demodulating, a neural network for outputting an output value of each output unit in accordance with the received light signal demodulated by the receiver, and a magnitude of an output value of each output unit output from the neural network. A comparison is made to determine the output value at which the output value is maximized. A tongue movement detection type syllable recognition device including a comparator that outputs a syllable corresponding to the force unit. 2. A microphone, wherein the syllable recognition unit detects a voice uttered during a learning operation and converts the voice into a detected voice signal.
An amplifier for amplifying the detected voice signal converted by the microphone; and a voice recognition circuit for recognizing a syllable from the detected voice signal amplified by the amplifier and outputting a teacher signal pattern indicating the syllable. The neural network further includes a learning function of learning a relationship between the received light signal and a syllable from the teacher signal pattern input from the voice recognition circuit and a received light signal demodulated by the receiver during a learning operation. Item 4. A syllable recognition device of the tongue movement detection type according to Item 1.